5

I have seen in tensorflow tutorials, they provide some interesting statistics about different architectures, such as number of operations taking place, etc .

This model achieves a peak performance of about 86% accuracy within a few hours of training time on a GPU. Please see below and the code for details. It consists of 1,068,298 learnable parameters and requires about 19.5M multiply-add operations to compute inference on a single image.

How can I achieve something similar using Caffe?
How can compute or compile such interesting statistics myself?
(by the way I'm not talking about tensorboard visualizations, Just statistics such as number of operations and stuff like that)

Rika
  • 19,296
  • 28
  • 91
  • 182

1 Answers1

1

There is an open issue #2507 in BVLC/caffe github requesting this feature.
You are more than welcome to make a contribution to caffe implementing this feature!

Shai
  • 93,148
  • 34
  • 197
  • 325
  • Thanks, I already created a script to calculate number of trainable parameters using deplot.prototxt. but I am clueless how to get FLOPs related statistics! – Rika Feb 02 '17 at 08:46